Credits
What credits buy.
Credits are the one unit for everything your agents spend: model work, web research and computer time, all from one balance. 1,000 credits is $1, whether they come with your plan or from a top-up.
Model work
Charged per million tokens. You never count tokens yourself: every task shows what it consumed when it finishes.
| Model | Input / M | Cached input / M | Output / M |
|---|---|---|---|
| GPT-5.6 LunaFast, the default | 500 | 50 | 3,000 |
| GPT-5.6 TerraBalanced | 5,000 | 500 | 30,000 |
| GPT-5.6 SolThe flagship | 10,000 | 1,000 | 50,000 |
Web research
Charged per operation.
| Operation | Credits |
|---|---|
| Read a page | 3 |
| Search, up to 10 results | 5 |
| Read a page that needs JavaScript | 11 |
| Read a page with a question or extraction | 11 |
| Search with page content, 10 results | 25 |
Computers
Computers run on demand and are charged per minute while running, plus a small charge to start, resume or suspend. A paused computer keeps its state and costs nothing to keep.
| Operation | Credits |
|---|---|
| Running | 4 / minute |
| Start or resume | 50 |
| Suspend | 100 |
What that means in practice
- A web search: about 5 credits
- A one-hour computer session, one start and one suspend: about 390 credits
- A substantial coding task: usually a few hundred credits, depending on context and model
Simulate a task
Pick a job and see where the credits go, priced with the rates above.
Research a market and write a brief
- 10 web searches50
- 15 page reads45
- 5 pages that need JavaScript55
- Model input, about 600k tokens (Luna)300
- Model output, about 40k tokens120
- Total, give or take570 credits
Roughly 17 times a month on Starter, 61 times a month on Pro, 210 times a month on Team.
These are launch rates and they are reviewed as provider prices change; your workspace always shows the current rates and exactly what each task consumed. Agent work varies with context, retries and tools, so treat the examples as typical, not promised.